Walter, I've found that with gcc I have a builtin.h header with several atomic functions.
There are functions for signed and unsigned 32bits increment/decrement, which one should I use? I see that in windows you cast to a (long *), but hvm.c passes to HB_ATOMIC_INC() a ulHolders, for example, line 9355. So, which is the correct operation to do? Last one, I still need to use mutexes at the .prg level, am I right? Maurilio. PS. I can add builtin.h if you need to look at it. Walter Negro wrote: > Maurilio, > > if you test mtcomplexshare.prg without the change in my commit, you can > obtain GPFs or Internal Error (Premature Pointer Release detected). > > The problem in low level is similar to problem in high level that you > see executing mtcomplexshare.prg without mutex. > > Some platforms or compilers do not have functions for atomic increment > and decrement, but some compilers create atomic increment and decrement > when using vars volatile. > > > > *Walter Negro* > /Desarrollo e Investigación/ > VsTour - Softmagic S.R.L. > <mailto:[EMAIL PROTECTED]> > ------------------------------------------------------------------------ > *::* * VSTour.COM <http://www.vstour.com> * *::* > > > Maurilio Longo escribió: >> Walter, >> >> I'm not sure I'm understanding this, but I did a test with >> mtcomplexshare.prg >> and I see that nGlobalCounter is protected by a mutex lock/unlock >> sequence. >> >> If I remove those mutex calls on OS/2 with GCC at the end the result >> shows big >> differences. >> >> --------8<---------------- >> Thread 1 - LocalCounter= 5449444 >> Thread 2 - LocalCounter= 5469999 >> Total 10919443 >> >> GlobalCounter= 6345520 >> >> Difference 4573923 >> (E:\repos2\xharbour\tests) >> -------->8---------------- >> >> See attachment. >> >> Maurilio. >> >> >> >> Walter Negro wrote: >> >>> 2008-06-28 15:25 UTC-0300 Walter Negro <anegro/at/overnet.com.ar> >>> * include\hbdefs.h >>> * include\hbvmpub.h >>> * include\thread.h >>> * source\rtl\hbxml.c >>> * source\vm\arrays.c >>> * source\vm\arrayshb.c >>> * source\vm\codebloc.c >>> * source\vm\fastitem.c >>> * source\vm\fm.c >>> * source\vm\garbage.c >>> * source\vm\hvm.c >>> * source\vm\itemapi.c >>> * source\vm\thread.c >>> * tests\mtcomplexshare.prg >>> * Added atomic or volatile increment and decrement of counters of >>> complex types and structures. >>> Now is safe to share complex data type in diferent threads if C >>> compiler or OS support >>> function or create ASM thread-safe to inc/dec. >>> NOTE: Atomic increment and decrement need counter memory aligned >>> and 32 bits of lenght. >>> >>> Walter Negro >>> >>> >>> ------------------------------------------------------------------------ >>> >>> >>> No virus found in this outgoing message. >>> Checked by AVG. Version: 8.0.101 / Virus Database: 270.4.2/1523 - >>> Release Date: 28/06/2008 07:00 a.m. >>> >>> >>> ------------------------------------------------------------------------ >>> >>> ------------------------------------------------------------------------- >>> >>> Check out the new SourceForge.net Marketplace. >>> It's the best place to buy or sell services for >>> just about anything Open Source. >>> http://sourceforge.net/services/buy/index.php >>> >>> >>> ------------------------------------------------------------------------ >>> >>> _______________________________________________ >>> xHarbour-developers mailing list >>> [email protected] >>> https://lists.sourceforge.net/lists/listinfo/xharbour-developers >>> >> >> > > > -- __________ | | | |__| Maurilio Longo |_|_|_|____| farmaconsult s.r.l. ------------------------------------------------------------------------- Check out the new SourceForge.net Marketplace. It's the best place to buy or sell services for just about anything Open Source. http://sourceforge.net/services/buy/index.php _______________________________________________ xHarbour-developers mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/xharbour-developers
